Context engineers are Agentic AI designers and system orchestrators
Instead of writing code, they write markdown notes to harness an AI Model.
An agentic harness is
A structure of markdown notes used to construct the AI’s 6 types of context;
- Instructions: Defining the agent’s core roles, goals, and boundaries.
- Knowledge: Connecting relevant domain-specific data, architectural diagrams, and retrieved documents.
- Memory: Managing short-term session logs (what just happened) and long-term persistent state (the project itself).
- Examples: Providing few-shot behavioral demonstrations and codebase reference patterns.
- Tools: Defining the APIs, scripts, skills and external services the agent is allowed to invoke.
- Guardrails: Enforcing constraints, formatting rules, and safety validations.
This system accounts for 90% of your AI capabilities and intelligence. The remaining 10% comes from the model you choose to place in your harness.
You can start learning for free by organizing your digital assets in Obsidian.